The present invention relates generally to chemical vapor deposition (CVD) processing, and more particularly to a method and apparatus for CVD chamber cleaning.
CVD is widely used in the semiconductor industry to deposit films of various kinds, such as intrinsic and doped amorphous silicon (a-Si), silicon oxide (SixOy), silicon nitride (SirNs), silicon oxynitride, and the like on a substrate. Modern semiconductor CVD processing is generally done in a vacuum chamber by using precursor gases which dissociate and react to form the desired film. In order to deposit films at low temperatures and relatively high deposition rates, a plasma can be formed from the precursor gases in the chamber during the deposition. Such processes are known as plasma-enhanced CVD processes or PECVD.
State of the art CVD semiconductor processing chambers are made of aluminum and include a support for the substrate and a port for entry of the required precursor gases. When a plasma is used, the gas inlet and/or the substrate support is connected to a source of power, such as a radio frequency (RF) power source. A vacuum pump is also connected to the chamber to control the pressure in the chamber and to remove the various gases and particulates generated during the deposition.
In all semiconductor processing, particulates in the chamber must be kept to a minimum. Particulates are formed because, during the deposition process, the film is deposited not only on the substrate, but also on walls and various fixtures, e.g., shields, the substrate support and the like in the chamber. During subsequent depositions, the film on the walls, etc., can crack or peel, causing contaminant particles to fall on the substrate. This causes problems and damage to particular devices on the substrate. Damaged devices have to be discarded.
When large glass substrates, e.g., of sizes up to 360 mm×450 mm or even larger, are processed to form thin film transistors for use as computer screens and the like, more than a million transistors may be formed on a single substrate. The presence of contaminates in the processing chamber is even more serious in this case, since the computer screen and the like will be inoperative if damaged by particulates. In this case, an entire large glass substrate may have to be discarded.
Thus, the CVD chamber must be periodically cleaned to remove accumulated films from prior depositions. Cleaning is generally done by passing an etch gas, particularly a fluorine-containing gas, such as nitrogen trifluoride (NF3), into the chamber. A standard method of performing this cleaning procedure is to pass a constant flow of NF3 into the chamber. A plasma is initiated from the fluorine-containing gas which reacts with coatings from prior depositions on the chamber walls and fixtures, e.g., coatings of a-Si, SixOy, SirNs, SiON and the like, as well as any other materials in the chamber. In particular, the NF3 creates free fluorine radicals “F*” which react with Si-containing residues.
The reaction forms gaseous fluorine-containing volatile products that can be pumped away through the chamber exhaust system. This procedure is generally followed by a nitrogen purge.
As the volatiles are pumped away, the F* contribution to the overall pressure stays low until there are no residues left with which the F* can react. Then the F* contribution to the overall pressure, and thus the overall pressure itself, rises. One can use this rise to detect the endpoint of the cleaning procedure by monitoring the overall pressure with a manometer and waiting for a user-defined endpoint pressure to be reached.
This technique is vulnerable to, e.g., variations in the manometer pressure reading over time, i.e., “manometer pressure drift”. Such variations are inherent in any such instrument and may be typically provided for with appropriate calibration techniques. If the variation is such that the manometer drifts low, the process may not seem to attain the endpoint. In this situation, the cleaning recipe is carried out indefinitely, or until a preset time is reached, termed a “recipe timeout”. Even though the chamber is cleaned, throughput is reduced and cost increased, the latter especially because NF3 is significantly more expensive than that of any other process gas. Flowing a large and constant amount of NF3 is costly and inefficient.
On the other hand, if the manometer drifts high, the preset pressure may be attained prior to endpoint, resulting in an incomplete clean. Residues may be left in the chamber that are a potential source of contaminant particles.
One way to conserve NF3 is to install a governor on the valve inlet of the NF3. The governor allows NF3 flow in such a manner as to keep the pressure of NF3 constant within the chamber. In this way, the flow rate is no longer constant, decreasing near the endpoint and saving costly NF3. A drawback of this solution is that the endpoint of the cleaning procedure cannot be detected using the rise of the NF3 level, since its pressure is by design constant.